Research
What was looked at before acting: the question, what was found, and what it rests on.
How did languages that were first said from outside - by assembly, by another language - come to say themselves, and what does that order teach for moving what the program says into the lenses?
In stages, never at once. A first translator is written outside - in another language, as an interpreter, for a subset, or by hand - and needs only to be enough to carry the language's own saying of itself; that saying is then run through it, and then through itself, and the two results are compared: where they differ, one of them is wrong. So the outside is not thrown away when the inside can first speak; it stays beside it until the inside says the same, and what remains outside afterwards is small and is what touches the machine. For ocoeco this gives the order: move what is already declared (what an entry asks, how a row is named, where it is kept) before what acts on a surface (reading the words, writing the row); run the lens beside the program's saying and compare before taking the program's away - which is what was done for entering, tried as a place's own lens first; and keep the kernel that stays in the program small and named - the engine that answers queries, reading from surfaces, writing rows. What the comparison is for a compiler, a look kept before and compared after is here.
2026-11-16
How is a moment said so that two of them can be compared by what they say alone?
RFC 3339 gives the form - date, T, time, and Z for UTC - and says in its section on ordering that where every time is in the same clock and the same form, the words sort as the times do; with offsets that differ they do not. Git keeps a change's moment with its maker's offset and offers it in the reader's own clock: --date=format-local, read with TZ=UTC, says it in UTC. The guard on the route holds it: a change made at 00:08:46+01:00 is said as 23:08:46Z of the day before.
2026-11-15
Has laying out a tree of data as text been said as rules in a language, and not written as a program?
Yes, for decades. XSLT is a language of rules that turn a source tree into a result tree - a stylesheet: what is laid out and how are kept apart, and the how is data that can be exchanged. Taken: the plane as what is given - ours is rows, in the order they are laid out; the surface as rules over it, in the language everything else is said in - a lens; and that one plane can have many surfaces with the program knowing none of them. Not taken: a second language for it.
- XSL Transformations (XSLT) Version 1.0 - W3C; James Clark, editor, 1999 Rules, said in a language of their own, that turn a source tree into a result tree; such a set of rules is called a stylesheet. What is laid out and how it is laid out are kept apart, and the how is data.
2026-11-15
What does one find, standing on a thing - looking it up by its name?
Linked Data's rules (Berners-Lee, 2006): when someone looks up a name, provide useful information about the thing - and include links to other things, so that more can be discovered. SPARQL's DESCRIBE form returns a description of a resource: what the service holds about it, in a form left to the service. Both give the thing's own description together with what it is linked to. The lens of around gave the links without the description.
2026-11-15
How is something that lives on a surface outside a place - a history in git, what runs on a machine, a table kept elsewhere - read as rows, and where is it said what a table is and where its rows come from?
Others who read a surface as a table agree on three things. The shape of a table is data: declared, not coded. The kinds of surface are few, and each is a small adapter in the program that makes rows when asked - nothing is stored. And which surface a table reads from is said beside its shape, the where a parameter. So here: a table's shape is said once - by the lens that defines it, else by the place's own declaration, else read off its rows; where its rows come from is a route, said in the language; and a table is entered as itself where no entry has its name. Three open things - a route said by a lens, a shape said where entries do not read, a next step that is words - were one question: what a table is, and where its rows come from.
- osquery: an operating system as relational tables - Facebook (now the osquery project), 2014 Each table is declared as data - its name, what it is, its columns - and its rows are made when asked for, by a small plugin that parses a file or calls the system: nothing is stored. The kinds of thing that can be read are plugins; what a table is, is a declaration.
- mergestat-lite (formerly askgit): git repositories as SQL tables - MergeStat, 2020 A commits table over a repository - hash, message, author, when - where which repository is a parameter of the table; the same for the files and the GitHub API. A history is read as rows where it is, not copied into a record of its own.
- Getting Things Done: the next action - David Allen, 2001 A project's next step is not a note on the project: it is an action of its own, concrete enough to be done and seen done. What is next is found among actions, not written as words on what they are for.
2026-11-14
Is there a sense in a monad in our ecosystem - can all lenses be expressed from a simplest, passthrough lens - and how does it fit with shapes coming down to bits?
Two things share the name monad and neither is the passthrough lens. In Haskell a monad chains steps that each carry something extra; what composes lenses there is plainer - optics compose as functions, and the identity is the lens that passes everything. Leibniz's monad is a standpoint: a simple substance mirroring the whole from its point of view - nearest to a place one looks from. What the world found as the smallest whole set: the identity; three ways into a shape - one part of what is side by side (a lens), one case of what is one of several (a prism), each of many (a traversal); two ways to go together - one after another, and side by side; and two directions - get and put - held to a law, that what is put is what is then got. Shapes themselves are made of the same two: this and that, this or that - down to the bit, the smallest this-or-that. A region of bits (from, how many) is the simplest lens there is, and formats are declared so. Against ocoeco (2026-11-14): both ways of going together exist (a shape read across lenses; places). The identity does not: a table is kept as another kind of thing than a lens, though it is the passthrough of what a place keeps. A lens's two faces, glimpse and focus, are put and get, but their law is nowhere said - one sign of it shows by accident: a table nothing brings into focus. And a lens is still a bundle of four things - what data is, how it enters, what bears on what, what comes into focus - with how it is laid out mixed in. Parts that count or gather only read: through those nothing can be put, and that is as it should be.
2026-11-14
How do tables serve the commons today, do they do something a graph of the best kind cannot, and how has the world dealt with what we are dealing with?
Looked at ourselves first (2026-11-13). Of 36 parts in the lenses built in: 14 only filter the rows of one kind; 15 count or gather; 9 bring two kinds together; 8 take a name apart to find what it points at; 1 is placed. All 22 tables are declared as a minimum - a row may carry more - and all are keyed by id alone: they are kinds of thing with an id, not tables in the strict sense. And the work about what bears on what is done outside any table: 280 lines of the program check links, 473 rewrite them when rows move, 59 walk a chain. So tables serve us for counting and gathering, for a whole thing readable on one line, and for checks when written; they do not give identity, edges, walking, what points at a thing, or what was. In the world: a table is one way of sorting facts, not the facts - Datomic keeps facts and answers row-like and column-like questions from the same ones; identity can be what a thing is, with names as labels (Unison); a graph can be declared over tables that exist and walked in the engine we use (SQL/PGQ); and text at rest with a graph when asked is a working arrangement (Logseq), whose newer form warns against giving up the text. In power a table does nothing a graph of facts cannot; in practice it gives a mature, fast engine and a row a person can read whole. The middle man is at the asking, not at rest: at rest we already keep things with ids and links - nodes and edges, written as text.
2026-11-13
Read as a database, how sound is ocoeco's data?
Measured 2026-11-12 across 22 projects: 14 tables; about 860 links between rows written as text (research 143, sources 317, decided 75, advances 88, touches 104, serves 129, answers 1), of which 2 point at an id found nowhere. Sound in practice, but held together by discipline: no relation is declared, so nothing checks one. Read with tad sql, each project's data already answers joins; what is missing is the shared language saying which field refers to which table, and a check across projects.
2026-11-12
Does open source have the intelligence and analysis a business has?
From the outside, yes: CHAOSS (the Linux Foundation, since 2017) defines metrics of activity and health, and GrimoireLab and Augur gather them from commits, issues, reviews and mailing lists. From the inside, no: why things are as they are, what was decided and rejected, what is needed next, who is doing it, and a way for the people who use the software to see that and take part. That inside is the analysis half of business intelligence and analysis; it is what ocoeco records as the work's own data, oco-agent coordinates, and the observatory presents.
- CHAOSS: Community Health Analytics in Open Source Software - The Linux Foundation, 2017 Metrics and tools (GrimoireLab, Augur) that measure open source projects' activity and health from the outside: commits, issues, reviews, mailing lists.
2026-11-12
What should the board be, taking the best the world has?
Kanban (Anderson, from Toyota's production system): make the flow visible, limit work in progress, pull, make policies explicit, measure time. Reinertsen: queues and ageing are the hidden cost. GitHub Projects and Linear: an item is kept once, with fields, and seen through many views - a board is a view, never a second list. Shape Up's hill charts: show uncertainty, not only state. For OCO: the board is a view over every project's backlog - one truth, the data the work already keeps - with the flow open, next, in-progress, waiting, parked, done, dropped; each item in progress held by a driver, with its next step and the day it entered its state; and what needs attention said plainly: a driver holding too much, work with no next step, work that has not moved for a week. oco-agent speaks it; anyone - a person or a machine - pulls from next. Not yet: uncertainty (hill charts), and what someone outside could pick up.
- Kanban: Successful Evolutionary Change for Your Technology Business - David J. Anderson, 2010 Make the flow of work visible, limit work in progress, pull rather than push, make policies explicit, and measure how long work takes.
- The Principles of Product Development Flow - Donald G. Reinertsen, 2009 Queues are the hidden cost of development; work that waits and ages is the signal to watch, more than how busy people are.
- About Projects - GitHub, 2025 Items kept once, with fields of one's own, and many views - table, board, roadmap - over the same items.
- The Linear Method - Linear, 2023 Work as issues with an owner and a state; a small set of states; triage for what comes in; momentum over ceremony.
- Shape Up, chapter 13: Show Progress (hill charts) - Ryan Singer, Basecamp, 2019 Progress as uncertainty: uphill while still figuring it out, downhill once it is only doing.
2026-11-12
Lens by lens, what does the best of the world's experience say - stripped of ceremony - and what does ocoeco keep or lack for it?
PURPOSE - best: say why a thing exists and what it must reach in terms that can be checked, and few of them. Keeps: the idea, milestones with criteria and readiness. Lacks: nothing essential. KNOWLEDGE - best: every claim with its source, and a question before an answer. Keeps: research, sources, concepts, vocabulary. Lacks: how a finding is known - measured, read, or reasoned; on 2026-11-12 a reasoned guess (a request with no time limit) stood beside measured facts and was wrong. REASONING - best: a decision on one page with its context, what was rejected, its consequences and its status (Toyota's A3, Nygard's decision records). Keeps: decisions with why, rejected, research, sources. Lacks: the relation between decisions - one refining or replacing another is written as text today, three times on 2026-11-12. WORK - best: visible flow, pull, limits, ageing (Kanban, Reinertsen). Keeps: the backlog's flow, driver, next, since, waits_on; oco-agent's journal says what actually ran. Lacks: why this work - its link to the milestone or decision it serves. RESULTS - best: evidence that can be checked, not a claim. Keeps: glimpses, workings with evidence, gathered releases, activity and test runs. Lacks: evidence as a link to the thing that proves it. LOOKING BACK - best: the After Action Review, hansei. Keeps: reflections. Lacks: nothing essential. PROBLEMS - best: the andon - a problem visible the moment it happens, to everyone, until it is dealt with. Keeps: nothing of its own; attention is computed for the flow only. Lacks: the lens itself - 21 release runs failed unseen and a watcher was silent for a night on 2026-11-12. The largest gap. INVITATION - best: a need marked where it is, small and well-defined, fillable by whoever passes (Wikipedia's citation needed, good-first-issue labels, citizen science, device owners filling a support table). Keeps: nothing. Lacks: the entity - what is needed, of what kind (information, a test on a device, feedback), where, and how to give it. What open commons add to Toyota's room. THE SKY ITSELF - what makes lenses one dome: relations seen as connections, the story, and time. Keeps: about 860 links that resolve, the essence a star tells, and history in git. Lacks: the relations declared - the dome's continuity rests on them. So, in order: declare the relations (among them one decision refining another, and work serving a milestone); the problems lens; the invitation; how a finding is known and evidence that links; then lenses as data.
- Obeya - Lean Enterprise Institute, 2024 A big room (Toyota, the Prius project, Takeshi Uchiyamada, 1990s) whose walls show timing, milestones, progress and countermeasures to problems together, so that problems become visible; begun because no one person held the knowledge the work needed.
- Kanban: Successful Evolutionary Change for Your Technology Business - David J. Anderson, 2010 Make the flow of work visible, limit work in progress, pull rather than push, make policies explicit, and measure how long work takes.
- The Principles of Product Development Flow - Donald G. Reinertsen, 2009 Queues are the hidden cost of development; work that waits and ages is the signal to watch, more than how busy people are.
- Documenting Architecture Decisions - Michael Nygard, 2011 A decision as a short record: its context, the decision, its consequences, and its status - proposed, accepted, superseded by another.
- Toyota Production System - Toyota Motor Corporation, 2024 Jidoka and the andon: a problem is made visible the moment it happens, and the work stops for it rather than passing it on.
- Wikipedia: Citation needed - Wikipedia, 2024 A need marked exactly where it is, in the thing itself, so that whoever passes can fill it.
- Zooniverse: people-powered research - The Citizen Science Alliance, 2024 Large work done by many through small, well-defined contributions that need no standing in the project.
- A Leader's Guide to After-Action Reviews (TC 25-20) - US Army, 1993 Four questions after any action: what was planned, what happened, why, and what to do next time.
- Governing the Commons: The Evolution of Institutions for Collective Action - Elinor Ostrom, 1990 Commons last where those who use a resource can see its state and take part in its rules: clear boundaries, monitoring by the users themselves, collective choice, and enterprises nested within larger ones.
2026-11-12
What is a board, taking the best of the world's experience and leaving its ceremony out - and what is it for open commons?
The best of it is Toyota's obeya, the big room: begun on the Prius project because no single person held the knowledge the work needed, its walls showed purpose, plan, progress, problems and countermeasures together, so that anyone entering saw the whole and problems became visible. Around it: make things visible (mieruka), signal a problem at once (andon), go and see the real thing, pull work rather than push it, a decision on one page with its reasoning (A3), and look back (hansei). What was added later - command centres, score boards, points, ceremonies - is the part to leave out. A board, then, is not a list of tasks: it is the room, and a flow of work is one of its walls. ocoeco's entities already are those walls: the idea and milestones (purpose); research, sources, concepts and words (knowledge); decisions with what was rejected (reasoning); the backlog's flow (work); glimpses, workings and their evidence (results); reflections (looking back); what needs attention (problems). Open commons need one wall Toyota's room never did: there nobody is assigned, so the room must also invite - show what is needed from whoever comes: a piece of information, a test on a device, feedback. Ostrom found commons last where those who use a thing can see its state and take part in its rules.
- Obeya - Lean Enterprise Institute, 2024 A big room (Toyota, the Prius project, Takeshi Uchiyamada, 1990s) whose walls show timing, milestones, progress and countermeasures to problems together, so that problems become visible; begun because no one person held the knowledge the work needed.
- Governing the Commons: The Evolution of Institutions for Collective Action - Elinor Ostrom, 1990 Commons last where those who use a resource can see its state and take part in its rules: clear boundaries, monitoring by the users themselves, collective choice, and enterprises nested within larger ones.
- Kanban: Successful Evolutionary Change for Your Technology Business - David J. Anderson, 2010 Make the flow of work visible, limit work in progress, pull rather than push, make policies explicit, and measure how long work takes.
- The Principles of Product Development Flow - Donald G. Reinertsen, 2009 Queues are the hidden cost of development; work that waits and ages is the signal to watch, more than how busy people are.
2026-11-12
What does a lens do with what passes through it - what is its second face called?
The author, 2026-11-12: a lens propagates light, it lets it through; by itself it shows nothing - it is not a screen or a picture - and the word showing falls off the full picture, as if from another domain. In optics a lens is defined by this: it transmits light and brings it to a focus; a telescope's lens gathers a star's faint light and focuses it, so that a blur becomes something one can make out. So the second face of a lens is its focus, and the act is to focus. It is plain and can be guessed (focus the lens on a star); it says the true thing about the data - nothing is made, the same data is brought into focus; and it leaves the showing where it happens, on the fulldome. The whole then reads as one picture: a star gives off light, its data; a glimpse catches it where it is fresh; a lens lets it through and brings it into focus; the fulldome is where it is seen; to observe is to gather the light of many stars. Shine is no act of ocoeco: it is what a star does by working.
- Lens - Wikipedia, 2025 A lens is a transmissive device: it lets light through and, by bending it, brings it to a focus. It adds no light and displays nothing; what is seen forms where the light comes to focus.
- Fulldome - Wikipedia, 2025 Immersive projection filling a dome - a planetarium's whole sky around the viewer - used to show not only the stars but their stories, and navigated in real time in digital planetariums.
2026-11-12
What should a look back over a stretch of work be called, and what should it hold?
The author's word, 2026-11-12, is reflection. Schon showed practitioners learn by reflecting on action; the US Army's After Action Review asks what was planned, what happened, why, and what next; Agile retrospectives are a team ceremony at an iteration's end; blameless postmortems are written after failures. ocoeco needs the plain, general one: anyone - a person or a machine - looking back over any stretch, at any time, about success or failure, saying what they saw, what they learned and what changes.
- The Reflective Practitioner: How Professionals Think in Action - Donald A. Schon, 1983 Practitioners learn by reflecting in action and on action: looking back is part of the work, not apart from it.
- A Leader's Guide to After-Action Reviews (TC 25-20) - US Army, 1993 Four questions after any action: what was planned, what happened, why, and what to do next time.
- Agile Retrospectives: Making Good Teams Great - Esther Derby and Diana Larsen, 2006 A team looks back at the end of an iteration, at a set time - a ceremony of the iteration.
- Postmortem Culture: Learning from Failure (Site Reliability Engineering, ch. 15) - John Lunney and Sue Lueder, Google, 2016 After a failure, a written, blameless account of what happened and what changes.
2026-11-12
What should ocoeco's three acts be called?
The author proposed, 2026-11-12: pulse for record (a pulsating star), shine for show, observe for observe. Observe is plain, belongs with star and observatory, and replaces harvest and discover - a farm's word and a search's. Pulse is taken: since 2026-11-12 it means oco-agent's sign of life to the switch outside, and beat a driver's; and someone typing it could not guess it writes a row, where record can be guessed. Shine reads two ways: in everyday speech a star that shines can be seen (show); in the sky what a star shines is its light, which here is its data (record). A word that can be read as either of two acts will be. Better to touch the act first - build one lens, run it - and name it then. Resolved the same day by the author: the act called record already had its word - glimpse (2026-11-11). A glimpse is the part of a lens that takes data in; so the three acts are glimpse, observe and show.
- Fulldome - Wikipedia, 2025 Immersive projection filling a dome - a planetarium's whole sky around the viewer - used to show not only the stars but their stories, and navigated in real time in digital planetariums.
2026-11-12
How should the commons choose its words?
Domain-driven design keeps one ubiquitous language used the same way in talk, code and documents - a word with two meanings is a bug. Orwell and the plain-language guidelines ask for the short everyday word a reader already knows, one term for one thing, and no jargon where an everyday word will do. So: look at what others call the thing and what the reader knows, prefer the plain word, keep one meaning, and record the words rejected.
- Domain-Driven Design (the ubiquitous language) - Eric Evans, 2003 One language, used the same way in conversation, in code and in documents; a word that means two things is a bug.
- Politics and the English Language - George Orwell, 1946 Prefer the short, everyday word; cut every word that can go; no jargon where an everyday word will do.
- Federal Plain Language Guidelines - Plain Language Action and Information Network, 2011 Write for the reader, in words they already know; one term for one thing, used the same way throughout.
2026-11-11
How should code be changed so an edit cannot silently miss?
Text patches break on formatting: nineteen times a pattern written for one layout met another. Structural tools (ast-grep on tree-sitter, Comby, rust-analyzer's replace) match code as code. Tried on a real bug: ast-grep matched a whole function by its signature and replaced it; it did not see an expression inside a write! macro, because macro insides are tokens. So: replace whole items, keep logic outside macros, put new views in new files, and let every edit stop the job when it does not take.
- ast-grep: structural search and rewrite - Herrington Darkholme, 2022 Code found and rewritten by its structure, written as code; formatting no longer matters. It cannot see inside Rust macros - their insides are tokens.
- Tree-sitter - Max Brunsfeld, 2018 An incremental parser behind ast-grep and many editors: the structure of code as a tree.
- Comby: structural code search and replace - Rijnard van Tonder, 2019 Structural rewriting by balanced delimiters, language-agnostic; another way, weighed and not taken.
2026-11-11
How do people measure repetitive manual work and how draining it is, and decide when to automate it?
Site reliability engineering names it toil - manual, repetitive, automatable, without enduring value - and keeps it within a budget. Perceived exertion is rated on one 0-10 scale (Borg CR10); workload can be split into six parts (NASA-TLX), too heavy for every record. Whether automation pays is time per occurrence times frequency (xkcd 1205). Toyota's jidoka makes the machine judge and stop itself, so no person stands by to judge.
- Eliminating Toil (Site Reliability Engineering, chapter 5) - Vivek Rau; ed. Beyer, Jones, Petoff, Murphy, 2016 Toil: manual, repetitive, automatable, tactical, without enduring value, growing with scale - counted and kept within a budget. The ledger counts exactly this.
- Psychophysical bases of perceived exertion - Gunnar Borg, 1982 A 0-10 category-ratio scale of perceived exertion: one number for how draining something was. The ledger's load.
- Development of NASA-TLX (Task Load Index) - Sandra G. Hart, Lowell E. Staveland, 1988 Workload in six parts - mental, physical and time demand, performance, effort, frustration. Too heavy for every bump; kept in mind for a richer load later.
- Is It Worth the Time? - Randall Munroe, 2013 How much time an automation may cost before it pays: time per occurrence times frequency over a horizon.
- Toyota Production System: Beyond Large-Scale Production - Taiichi Ohno, 1988 Jidoka, automation with a human touch: the machine judges and stops itself on anything abnormal, so no person stands by to judge. Why tests must judge sound and pictures themselves.
2026-11-11
How should the data about work be produced so it cannot go stale, and can ocoeco take over what release-please and release-plz do?
Measured 2026-11-11: only 224 of 464 recent landed commits across 22 projects (48%) changed any ocoeco data - half of all work leaves no trace in the system. 7 projects use release-please or release-plz; what anything consumes from them is the version and the tag (the conveyor builds binaries from tags, update-pins reads them, crates.io publishing hangs off them); their changelogs (23 to 470 lines of commit titles) drive nothing. Prior art: changesets and Knope record a change's intent when it is made and consume it at release (highest bump wins), but their record is only a bump and a summary, tied to nothing else. Git documents that local hooks can be bypassed (--no-verify); the guarantee has to be at the boundary that cannot be skipped - CI on the pull request, and land.
- Changesets - Atlassian, Thinkmill, 2019 Record the intent of a change when it is made, as a file beside it; a release consumes them, takes the highest bump, writes the changelog.
- Knope - Dylan Anthony, 2022 Changesets and conventional commits in one Rust tool: the version and changelog picked from both.
- githooks - hooks used by Git - Git project, 2005 Local hooks (pre-commit, prepare-commit-msg, commit-msg) can be bypassed with --no-verify; core.hooksPath moves them.
- git interpret-trailers - Git project, 2014 Machine-readable key: value lines at the end of a commit message (Signed-off-by, Fixes).
- release-please - Google, 2019 A release pull request kept from conventional commits: the version and a changelog of commit titles.
- release-plz - Marco Ieni, 2022 release-please's idea for Rust crates, with crates.io publishing.
2026-11-11
What should one recorded look at work be called, in the observatory's language?
Observatories keep an observing log; each look is an observation, recorded with its time, target, observer and notes, written at the eyepiece while fresh, and entries point to what they relate to. Weighed: snapshot (already a filesystem and VM word in the author's world - two meanings), exposure (also risk), frame (overloaded in software), observation (exact but long, and it blurs with what the observatory itself does), sighting (seeing someone else's object), entry (says nothing). glimpse: plain, short, one meaning here, and honest that each look is partial - together they make the picture.
- What I Saw When I Wasn't Sleeping: An Observing Log Primer - Tom Campbell, 2025 An observer records notes at the eyepiece, while it is fresh; the log becomes the record of the journey, one object at a time.
- OPSLOG: introduction to log entries - Green Bank Observatory (NRAO), 2004 Each log entry has a key, a time and a type fixed when it is made, and points to the entry it relates to.
2026-11-11
Other sources
- Bootstrapping (compilers) - Wikipedia, 2026 A compiler written in the language it compiles is reached in stages - a first one written in another language or by hand, enough to translate its own source; the full one built by it; then the full one built by itself. The full one is built twice so that the two can be compared: where they differ, the first or the full one is wrong. The first may be an interpreter, a subset the existing tools handle, or a translation by hand.
- The Thing (the fence) - G. K. Chesterton, 1929 Do not remove a fence until you know why it was put up: find out why something is as it is before changing it.
- Datomic: the data model - datoms - Cognitect / Nubank, 2012 A database is a set of immutable facts, each relating an entity, an attribute and a value, with the transaction that added it and whether it was added or taken back. There are no tables: any entity can have any attribute, and attributes are themselves entities. Nothing is updated in place. The same facts, kept sorted four ways, answer a row-like question (by entity), a column-like one (by attribute), a lookup by value, and what points at this (by the value referred to). A database is a value at a point in time; Datalog asks of it, with joins and recursion.
- The Toyota Way (go and see) - Jeffrey K. Liker, 2004 Go and see the real thing for yourself - the device's own record, not a theory about it.
- git-log: the forms of --date - The Git project The -local forms of --date say a change's moment in the reader's own clock, not its maker's: format-local with a form of one's own, read with TZ=UTC, says it in UTC.
- Kaitai Struct: a declarative language for binary formats - Kaitai project, 2016 A binary format is declared once as a sequence of fields - each with a kind and a size, down to single bits - with types nested in types, repeats and conditions; reading it is generated from the declaration. A shape of shapes over bits: regions side by side, regions within regions.
- The Monadology - G. W. Leibniz, 1714 A monad is a simple substance without parts; each has relations that express all the others, and so is a perpetual living mirror of the universe from its own point of view - as one town seen from different sides appears as many, which are only perspectives on a single one. Not a way of composing: a standpoint. In ocoeco it is nearest to a place one looks from, not to the simplest lens.
- Combinators for bidirectional tree transformations: a linguistic approach to the view-update problem - Foster, Greenwald, Moore, Pierce, Schmitt, 2007 A lens is two functions kept together, get (a view from a source) and put (a changed view back into the source), held to laws: what is put is what is then got (PutGet), and putting back what was got changes nothing (GetPut). It answers the databases' old question of how a change made to a view reaches what it is a view of. Small lenses are composed into larger ones by combinators that keep the laws, so the whole is right by how it is built.
- Linked Data - Design Issues - Tim Berners-Lee, 2006 Four rules: name things; so that the names can be looked up; when someone looks up a name, provide useful information; and include links to other names, so that more things can be discovered.
- Logseq: plain text at rest, a graph of facts in memory - Logseq, 2020 Notes are kept as plain text files; when read they are parsed into DataScript, a database of facts in memory asked in Datalog, and what is shown is asked again when the facts change. The arrangement tad has - text at rest, an engine when asked - with a graph in place of tables. A caution from the same project: its newer form keeps everything in a database file instead, faster for large graphs, and gives up files a person can read and compare line by line.
- The OODA loop - John Boyd, 1976 Observe and orient before deciding and acting; the looking is part of the work, not a delay before it.
- Optics in Haskell: lenses compose as functions, and the identity is a lens - lens-family (R. O'Connor) and the lens libraries, 2012 Lenses are composed with ordinary function composition, and the identity function is the identity lens: looking through it gives the thing itself. A lens focuses on one part of a thing made of parts side by side; a prism, its dual, on one case of a thing that is one of several; a traversal on many at once; a getter or a fold only reads. One family, composed one after another by the same operator. This is not a monad: a monad chains steps that each carry something extra, and is another structure.
- Ratatui - the Ratatui developers, 2023 Rust's terminal UI library, with a test backend that lets an interface be tested without a terminal.
- RFC 3339: Date and Time on the Internet: Timestamps - G. Klyne, C. Newman, 2002 Section 5.1, ordering: where every time is in the same clock - all in UTC - and in the same form, the times sort as words, and the order of the words is the order of time.
- SPARQL 1.1 Query Language: DESCRIBE - W3C; Steve Harris, Andy Seaborne, editors, 2013 The DESCRIBE form returns a description of the resources found: what the service holds about them - which properties, in which form, is left to the service.
- SQL/PGQ (SQL:2023) and DuckPGQ: graphs seen in tables - ten Wolde, Szarnyas, Boncz (CWI), 2023 The SQL standard of 2023 lets a graph be declared over tables that already exist - which are its nodes, which its edges - and then asked by pattern and by path, within an ordinary query. DuckPGQ brings it to DuckDB as a community extension; the graph is read, not written through. A way to walk what tables hold without leaving the engine ocoeco already uses.
- Unison: the big idea - content-addressed code - Unison Computing, 2019 Every definition is identified by a hash of what it is; names are separately kept labels pointing at hashes. A rename is changing which name points at a hash - instant, and nothing else has to agree. Conflicts between things of one name do not arise, since nothing is referred to by name; where two things share a name they are shown with part of their hash. Its own image: definitions are like stars in the sky.